...
Casablanca, Citizen Kane, The Third Man — these films exist only in black and white. AI colorization brings them into color, but the choices about what colors to use are creative decisions, not technical ones.
You watch the final scene of Casablanca. Rick and Captain Renault walk into the fog toward the Free French garrison. The film is black and white. The fog is gray. Rick's trench coat is gray. The runway is gray. Everything is a shade of gray — not because the world was gray in 1942, but because the film stock could only capture luminance, not color.
You run a frame from the scene through an AI colorizer. The fog becomes a pale, cool white. Rick's trench coat becomes beige. The runway becomes dark asphalt. The skin tones become warm and lifelike. The scene transforms from a historical artifact into something that feels almost contemporary — as if you are watching a modern film set in the 1940s, not a film made in the 1940s.
AI colorization of classic film stills is controversial among film historians — and fascinating to everyone else. Here is what the AI gets right, what it gets wrong, and why the color choices are creative decisions, not technical ones.
An AI colorizer trained on millions of color photographs can accurately predict: skin tones (the AI knows the range of human skin colors and applies them convincingly), sky and vegetation (sky is blue, grass is green, trees are green-brown — these are the most statistically reliable predictions), common materials (wood is brown, concrete is gray, brick is red-brown), and lighting conditions (the AI recognizes the difference between daylight, tungsten, and shadow and adjusts color temperature accordingly).
For a casual viewer, the AI-colorized film still looks convincing. The colors are plausible. The scene looks like it could have been shot in color. The AI has done its job — it produced a statistically probable colorization of a black-and-white image.
The AI does not know: the costume designer's intent (was Ingrid Bergman's dress in that scene blue or green? The AI guesses. The costume designer knew.), the set decorator's color palette (the furniture, the wallpaper, the props — these were chosen for specific visual effects that the AI cannot reconstruct), and the cinematographer's lighting choices (the lighting was designed for black and white — high contrast, deep shadows, dramatic highlights. Adding color changes the emotional impact of the lighting.)
Film historians object to AI colorization because it replaces the filmmakers' intentional creative choices with statistical guesses. The black and white cinematography of Casablanca was not a technical limitation — it was an artistic choice. The film was made in 1942. Color film existed. The filmmakers chose black and white for its dramatic effect. Colorizing the film changes the artistic intent, not just the pixel values.
AI colorization is appropriate for: personal family photos and home movies (these were shot in black and white for cost reasons, not artistic reasons — colorizing them restores information that was lost, not art that was intended), historical documentary footage (the goal is to make the past feel real and immediate, and color helps achieve that goal), and educational purposes (showing students what historical events might have looked like in color).
AI colorization is controversial for: artistically significant films (where the black and white cinematography was a deliberate creative choice), and any context where the colorized version is presented as historically accurate (the colors are guesses, not facts).
The AI colorizer is a tool of imagination, not reconstruction. It shows you what the past might have looked like. It does not show you what the past actually looked like. The difference matters.